- Emily R. Carrier1,*,
- James D. Reschovsky2,
- David A. Katz3 and
- Michelle M. Mello4
1Emily R. Carrier (firstname.lastname@example.org) is a senior health researcher at the Center for Studying Health System Change, in Washington, D.C.
2James D. Reschovsky is a senior health researcher at the Center for Studying Health System Change.
3David A. Katz is an associate professor in the Department of Internal Medicine at the University of Iowa, Iowa City, and a core investigator in the Comprehensive Access and Delivery Research and Evaluation Center, Veterans Affairs Iowa City Health Care System.
4Michelle M. Mello is the C. Boyden Gray Associate Professor of Health Policy and Law in the Department of Health Policy and Management, Harvard School of Public Health, in Boston, Massachusetts.
- ↵*Corresponding author
Despite widespread agreement that physicians who practice defensive medicine drive up health care costs, the extent to which defensive medicine increases costs is unclear. The differences in findings to date stem in part from the use of two distinct approaches for assessing physicians’ perceived malpractice risk. In this study we used an alternative strategy: We linked physicians’ responses regarding their levels of malpractice concern as reported in the 2008 Health Tracking Physician Survey to Medicare Parts A and B claims for the patients they treated during the study period, 2007–09. We found that physicians who reported a high level of malpractice concern were most likely to engage in practices that would be considered defensive when diagnosing patients who visited their offices with new complaints of chest pain, headache, or lower back pain. No consistent relationship was seen, however, when state-level indicators of malpractice risk replaced self-rated concern. Reducing defensive medicine may require approaches focused on physicians’ perceptions of legal risk and the underlying factors driving those perceptions.
There is widespread consensus that physicians alter their behavior to reduce their malpractice liability, but there is considerable disagreement about the magnitude of costs, harms, and potential benefits associated with the practice of such defensive medicine.1 Prior studies have produced conflicting findings, creating confusion for policy makers.2
In this article we explain why existing studies have not been able to provide conclusive information about the extent of defensive medicine. We also present a new analysis that combines the greatest strengths of the available approaches while avoiding some of their weaknesses. Linking physicians’ survey responses with their patients’ Medicare claims, we examined whether physicians’ self-reported level of malpractice concern was associated with their patients’ use of diagnostic imaging services and other tests, emergency department (ED) visits, and hospital admissions.
Our analysis examined assurance behaviors (sometimes called positive defensive medicine), which include a physician’s ordering additional tests, procedures, or visits and reducing his or her threshold for hospitalizing patients or referring them to other physicians. Another type of defensive medicine is avoidance (or negative) behaviors—such as physicians’ taking steps to avoid treating patients or conditions perceived to represent high legal risk.3 Although avoidance behaviors have troubling implications for patients’ access to care, policy makers tend to be more concerned about assurance behaviors because they contribute to escalating national health care costs.2
Why Measuring Defensive Medicine Is So Hard
This study’s approach—analyzing physicians’ concerns about malpractice liability as a predictor of their patients’ fee-for-service Medicare claims—blends the strengths of two standard approaches: opinion surveys of individual clinicians and cross-state comparisons of claims data over time. Opinion surveys typically query physicians about their assurance behaviors when presented with specific clinical vignettes or scenarios; researchers subsequently examine correlations between the physicians’ self-described decisions in these hypothetical scenarios and various measures of liability risk, usually malpractice insurance premiums or tort-reform laws in a state.3⇓–5 In contrast, we used a validated survey instrument included in a national survey of physicians who provided direct patient care,6 and we linked the physicians’ responses regarding their levels of malpractice concern to Medicare claims for the patients they treated during the study period.
Comparing claims for physicians who report high levels of concern about malpractice to claims for those who report less concern provides a better, more objective comparison of the two groups’ practice behavior in response to the malpractice environment than using the results of opinion surveys alone. On surveys, physicians may sometimes exaggerate their responses to questions that assess liability pressure, possibly to help make the case for malpractice policy reforms. Alternatively, physicians may be unable to discriminate between the effects of liability pressure and those of other factors that influence their decisions to order extra services, such as patients’ expectations and reimbursement incentives.
Linking claims data to a physician survey also avoids an important limitation of many claims-based studies: reliance on the absence or presence of tort reforms or high malpractice insurance premiums in the state where a particular physician practices as a proxy for that physician’s propensity to practice defensively. Studies using this approach have found small effects of defensive medicine on costs, typically less than 2 percent of total spending.7⇓–9 Studies that focus on specific high-risk specialties and patients have found larger effects.10,11 It is difficult to definitively conclude that differences in care use across geographic areas are attributable to differences in liability pressure, as opposed to other factors.
Additionally, these studies rarely incorporate direct measures of an individual physician’s personal assessment of his or her malpractice risk, relying instead on state-level measures. In earlier work we found state measures of malpractice risk and specific malpractice tort laws to be only weakly associated with physicians’ fears of malpractice suits,6 providing a possible explanation for the small estimates of the cost of defensive medicine in claims-based studies.
We hypothesized that physicians with high levels of malpractice concern would be more likely than others to order tests and procedures or refer patients to other providers to rule out the most worrisome potential diagnoses, all other factors being held equal. Of note, we would not expect malpractice-sensitive physicians to use more of all types of services for patients with all symptoms, because some services play little role in the diagnosis of potentially worrisome conditions. Instead, we would expect those physicians to increase their use of specific services to either diagnose or rule out severe conditions as causes of the symptoms in question, without changing their use of other services.
For reasons described below, we chose to examine the use of services for patients visiting physicians with complaints of chest pain, headache, or lower back pain.
For chest pain patients not initially seen in the ED, we expected malpractice-sensitive physicians to use more conventional and more advanced imaging and to be more likely to have their patients visit the ED or be admitted to the hospital, compared to their less sensitive peers. We expected that the use of stress testing might be inversely associated with malpractice concern, if physicians with lower levels of concern were more likely to use this test to assess patients’ risk in lieu of hospitalizing them. For headache, we expected malpractice-sensitive physicians to use more advanced imaging and ED referrals, but not more hospital admissions or conventional imaging. For lower back pain, we expected malpractice-sensitive physicians to use more conventional imaging and more advanced imaging, and possibly more ED referrals, but not more hospital admissions.
We found that higher levels of malpractice concern were indeed associated with significantly higher use of diagnostic services when the comparison employed self-reported levels of malpractice concern. However, that relationship was not observed when state-level measures of risk were used instead. This finding suggests that comparisons between states that have differing malpractice environments might not detect the presence of defensive medicine.
Study Data And Methods
We used a national sample of elderly Medicare beneficiaries enrolled in traditional Medicare, linked to nationally representative physician data from the 2008 Center for Studying Health System Change Health Tracking Physician Survey (which had a response rate of 62 percent).12 The characteristics of this physician sample—which asked respondents for demographic information, specialty, and practice characteristics and about their concerns regarding involvement in malpractice suits—have been previously reported.6 Our beneficiary sample included nearly 1.9 million patients who received services from 3,469 physician survey respondents in the period 2007–09. For these linked beneficiaries, full Parts A and B claims were obtained for the same period. Survey sampling weights were applied to the beneficiary sample to make it nationally representative.13
Measures Of Physicians’ Malpractice Concerns
The Health Tracking Physician Survey included five previously validated questions regarding physicians’ fears of malpractice lawsuits and their propensity to engage in positive defensive medicine.14⇓–16 These questions are shown in online Appendix Table A1.17 Each question elicits physicians’ responses on a five-point Likert scale that ranges from “strongly agree” to “strongly disagree.” We calculated physicians’ overall level of malpractice concern by summing the scores for the five questions, and we then classified physicians into terciles based on their total scores. An alternative specification based on questions directly focused on defensive medicine yielded similar results (see Appendix Table A2). Further information about the physician sample is available in Appendix Table A3.17
To further investigate which measures of malpractice risk best explain variations in the use of care, we compared findings from our survey-based index of malpractice concern to those from similar analyses based on state-level indicators of malpractice risk that have been used in prior claims-based comparisons. The first state variable was a risk index, defined as malpractice claims per 1,000 physicians times the average dollar amount of the award.18 The second was a binary variable indicating whether the state capped malpractice awards (noneconomic, punitive, or total damages). Caps on damages are often touted as a key mechanism for controlling defensive medicine.19 This step allowed a comparison between the estimates of defensive medicine based on a comparison of claims from different states and the estimates from our comparative approach.
Measures Of Defensive Medicine
We examined three measures of potentially defensive medicine that could occur either during an initial visit to a physician’s office—rather than the ED—for a worrisome complaint or within the week following the visit. Specifically, we noted whether the patient received certain diagnostic imaging services or other tests, was referred to the ED, or was admitted to a hospital. As previously noted, we analyzed patients who visited a physician with an initial complaint of one of the following three symptoms: chest pain, headache, or lower back pain. These three common symptoms were chosen because they represent a range of potential underlying problems, from mild and benign to severe and life threatening (see Appendix Figure A1 for details)17 and because physicians have considerable discretion in responding to these complaints.
Symptoms were indicated in claims using International Classification of Diseases, Ninth Revision (ICD-9), diagnostic codes. The initial visits we considered took place between July 1, 2008, and December 1, 2009. They were identified when claims data indicated that during the previous eighteen months, the patient had neither received any services for the same or a related diagnosis nor had a diagnosis for another condition that would justify the diagnostic test or service in question. Details on the included and excluded diagnoses are provided in Appendix Tables A4 and A5; unadjusted results are available in Appendix Tables A6 and A7.17 Results were insensitive to the length of the observation period following the initial visit.
Using logistic regression, we related service use during and after the initial visit to the physician’s level of malpractice concern, controlling for beneficiary and physician characteristics. Specifically, our models controlled for beneficiaries’ demographic characteristics,20 comorbid conditions,21 and urbanicity of residence, as well as the specialty of the physician at the initial visit. Full regression results are shown in the Appendix Tables A8–A25.17 Results are presented as adjusted percentages, by tercile of malpractice concern.
We stratified our analyses by whether the initial visit was to an ED or a physician’s office. We found that service use was dramatically higher when the initial visit was to an ED (see Appendix Table 6).17 For example, subsequent hospitalizations were fifteen to twenty-six times more likely for chest pain and lower back pain, respectively, which suggests that patients seen initially in EDs have very different clinical profiles than those who visit physicians’ offices instead. Moreover, diagnostic testing equipment is more readily available in the ED. In addition, emergency medicine physicians report higher rates of malpractice concern than other physicians do.6 However, malpractice concern was not consistently related to service use for patients whose initial visits were to an ED (see Appendix Tables A9, A11, and A13).17 For these reasons, we limit our presentation of results here to patients initially seen in physicians’ offices.
Differences in case-mix between comparison groups may not have been fully captured by the ICD-9 diagnostic codes. As a result, residual confounding could have biased our utilization results.
In addition, our sample included only patients who were diagnosed with a relatively benign cause of the presenting symptom and those for whom no underlying cause was identified (for example, chest pain of unknown origin). We excluded patients who were assigned a serious or life-threatening diagnosis. For example, patients with a diagnosis of nonspecific headache were included, but those diagnosed with subarachnoid hemorrhage were excluded. We excluded patients with serious, definitive diagnoses because we expected that they would have been seen by multiple other physicians whose level of malpractice concern was not measured.
Because physicians with high malpractice concern might be expected to conduct a more aggressive diagnostic evaluation initially, any bias created by this choice would tend to underestimate the use differences between physicians with high and low levels of malpractice concern. Appendix Figure A117 describes the range of health conditions that might be associated with the three symptoms and the types of tests used to diagnose these conditions in our analysis sample.
Beneficiary And Visit Characteristics
Of the nearly 1.9 million Medicare beneficiaries in our data set, 29,079 met the criterion of having an initial visit for one of the three symptoms in a physician’s office or an ED (Exhibit 1). The subsamples differed only modestly from each other in patients’ demographic and health characteristics and in characteristics of the state’s malpractice environment (Exhibit 1).17 Some beneficiaries with initial visits were missing data used in the statistical adjustment and thus are not represented in the adjusted samples.
Characteristics Of Medicare Beneficiaries And Their Visits To A Physician’s Office Or The Emergency Department For Treatment, By Symptom
When we focused on services provided to patients whose initial visit was to a physician’s office, we found that rates of hospitalization and referral to the ED were generally low (Exhibit 2). Rates of diagnostic tests were also low, with the exception of conventional imaging for chest pain and lower back pain, supplemental testing (stress tests) for chest pain, and advanced imaging for headache.
Unadjusted Percentages Of Patients Receiving Services During An Initial Visit To A Physician’s Office Or Within The Following 7 Days
Self-Reported Malpractice Concern And Defensive Medicine
Patients with chest pain had a significantly higher probability of being referred to the ED if their physician had a high or medium level of malpractice concern, compared to a low level (Exhibit 3; unadjusted percentages are shown in Appendix Table A6).17 Physicians with a high level of malpractice concern were significantly less likely than those with a low level to order stress tests, consistent with our hypothesis that they would be less inclined to use stress tests as a substitute for a more extensive inpatient workup. The hypothesized relationships between higher malpractice concern and higher rates of advanced imaging and hospitalization were observed but were not significant.
Adjusted Percentages Of Likelihood That A Visit To A Physician’s Office Resulted In Services During The Visit Or Within The Following 7 Days, By Physician’s Reported Level Of Malpractice Concern
Patients with headache who saw physicians with a high level of malpractice concern had a significantly higher probability of receiving advanced imaging than patients who saw physicians with a low level of concern—11.5 percent versus 6.4 percent (Exhibit 3). As expected, rates of conventional imaging and hospitalization were extremely low for both groups and were not significantly associated with the physician’s level of malpractice concern. Contrary to our expectations, there was no significant association between level of concern and referral to the ED.
Patients with lower back pain who saw physicians with a high level of malpractice concern had a significantly higher probability of receiving conventional and advanced imaging than patients seeing less-concerned physicians (Exhibit 3). There was no significant difference in their probability of hospitalization.
State-Level Malpractice Risk And Defensive Medicine
Only three significant associations emerged in our models analyzing state-level malpractice risk as a predictor of services received by Medicare beneficiaries (Exhibit 4). Two of these associations were in the opposite direction from those in the malpractice-concern model—that is, use was lower in states with a higher level of malpractice risk.
Adjusted Percentages Of Likelihood That A Visit To A Physician’s Office Resulted In Services During The Visit Or Within The Following 7 Days, By State Malpractice Risk Level
Models using the presence of caps on damages rather than malpractice risk as the liability measure indicated that caps were associated with the greater use of some services (Appendix Tables A20, A22, and A24).17 This counterintuitive finding might reflect a problem of reverse causality—that is, a high prevalence of malpractice concern (and defensive medicine) among physicians might make a state more likely to adopt a damages cap.
This study found that physicians who reported concern about their malpractice risk and who evaluated patients visiting their offices with chest pain, headache, or lower back pain were significantly more likely to order certain diagnostic tests, a pattern consistent with the practice of defensive medicine. However, these effects were not observed when state-level measures of malpractice risk were used in lieu of individual physicians’ reported concerns.
Compared to office-based physicians who reported a low level of concern about malpractice risk, physicians who reported higher levels of concern had higher rates of use of diagnostic imaging for lower back pain and headache. Although our analysis represents an initial effort to measure physicians’ behavior, physicians’ reported levels of concern appeared more strongly predictive of their use of services than were state-level measures of malpractice risk in the ambulatory care setting. Office-based physicians with higher levels of concern were more likely to order advanced imaging for patients with headache and lower back pain and conventional imaging for patients with chest pain and lower back pain, compared to physicians with low levels of concern. And compared to less-concerned physicians, more-concerned physicians were less likely to use stress testing for patients with chest pain and more likely to refer them to the ED, where a more extensive workup and observation would likely be performed.
This study provides insights into the magnitude of defensive medicine in the United States. Most previous studies have focused on the use of care in the hospital setting, especially obstetric care and the evaluation of patients in the ED. Our study provides a rare look at what happens in physicians’ offices.
In addition, previous studies usually focused on patients who constitute a relatively high liability risk. Malpractice awards in obstetrical cases, for example, can be enormous because they involve young patients and often result from catastrophic injuries. Patients who visit the ED for care are also risky to the diagnosing physician because their chances of having a serious problem are generally higher, on average, than patients who visit a physician’s office. In addition, ED physicians often lack detailed medical records for their patients and thus face greater uncertainty than office-based physicians. In contrast, our nationally representative sample was composed of elderly patients who were initially seen in a physician’s office. Although this is not a group that experts would identify as high risk from a liability perspective, the use of health services was significantly greater in this group of patients if they were seen by physicians with higher levels of concern about malpractice lawsuits.
Furthermore, our study helps reconcile the discrepancy between prior estimates of the cost of defensive medicine based on physician surveys and those based on state-to-state comparisons of defensive medicine. We found that associations between malpractice risk and use of health services were much stronger when the risk measure used was the physician’s perceived risk rather than objective, state-level measures of liability risk, such as claims costs or tort reforms. Our results suggest that prior cost estimates based on comparisons between states with and without caps on damages or high malpractice risk, malpractice premiums, or claims costs (including comparisons of variations in states over time) may not reflect physicians’ behavior or fully capture the influence of physicians’ malpractice concerns on defensive medicine.
Although it is impossible to say which approach captures the true prevalence of defensive medicine, our approach predicts individual patients’ use of services more effectively than do comparisons based on measures of the physician’s tort environment. Linking survey reports of physicians’ malpractice concern with claims data offers a promising way to generate more-accurate estimates of defensive medicine.
Lack Of Clear Association
We found no clear association between physicians’ malpractice concerns and the services their patients used when patients initially visited an ED instead of a physician’s office. One possible explanation for this finding is that emergency physicians have incorporated defensive practices into their routine care, so that even those who perceive themselves to have low levels of malpractice concern already practice in a defensive manner.22
In addition, we did not observe a consistent relationship between malpractice concern and the use of diagnostic imaging in patients with a new complaint of chest pain. This lack of association may indicate either that physicians do not practice defensively for patients with chest pain or that any defensive practices are obscured because a broader range of diagnostic approaches and the involvement of multiple clinical services may be required to rule out potentially serious causes of this complaint.
Finally, we did not observe a significant relationship between outpatient physicians’ malpractice concern and ED use by their patients with headache in the days following the physician’s initial evaluation. This may reflect the fact that patients with the most severe headache symptoms are more likely to go directly to an ED instead of visiting a physician’s office. Findings for lower back pain were more consistent with an association between physicians’ concern and patients’ ED use. Patients seen by a physician with a medium level of concern were significantly more likely to visit the ED than were patients seen by a physician with a low level of concern. Patients seen by a physician with a high level of concern were also more likely to visit the ED, but that result was not significant.
Overcoming Physicians’ Fear Of Malpractice Claims
Previous comparisons of states with and without tort reforms (such as caps on damages, limits on attorney’s fees, and reductions in the time that plaintiffs have to file a claim) suggest that the presence of tort reforms does little to limit use of health care. This has led to inferences that the total cost of defensive medicine is low.23 An alternative explanation, empirically supported by previous work, is that the kinds of tort reforms that states have adopted to date do not appreciably reduce physicians’ level of malpractice concern,6 and it is perceived rather than actual risk that determines how physicians behave.
Policy approaches that target the underlying causes of physicians’ malpractice concerns might reduce defensive medicine more effectively than current estimates suggest. Being sued is associated with substantial distress for physicians, but tort reforms are generally aimed at lowering the cost of eventual payouts to the exclusion of other approaches. Physicians’ extreme dread of malpractice litigation may stem from their perception that it is unpredictable, uncontrollable, and potentially disastrous both financially and psychologically.6
To some extent, this fear might be ameliorated simply by better informing physicians about the actual dynamics of the malpractice claims process. For example, prior studies show that physicians greatly overestimate their risk of being sued;24 the actual risk of facing a malpractice claim in specialties such as emergency medicine may not be any greater than the average risk for all physicians;25 only about one in fifty injuries caused by negligence result in a claim;26 lawsuits are virtually never settled for amounts in excess of a physician’s insurance policy limits (and thus involve little personal financial risk for physicians);27 claims are resolved in ways concordant with their merit three-quarters of the time;28 and malpractice plaintiffs lose four out of five trials.28
Physicians, most of whom are unaware of these data, may have exaggerated perceptions of legal risk. However, our prior work,6 which found that physicians’ levels of malpractice fear are not substantially different in states with and without tort reforms, suggests that physicians’ attitudes may be hard to align with their actual risk.
This persistence in risk perceptions may, again, reflect the dread involved in perceptions of this type of risk. The experience of being sued is undeniably distressing—it has been described as one of the most stressful experiences a physician can have,27 engendering shame, self-doubt, and emotional trauma.29⇓–31 A recent study found that the average physician spends almost 11 percent of his or her career practicing in the shadow of an unresolved malpractice claim.32 Neither insurance nor traditional tort reforms address these harms. Reforms need to reassure physicians that medical injuries can be resolved expeditiously and fairly, in a less adversarial manner.
To achieve this goal, reforms need to facilitate communication between physicians and patients about why adverse outcomes occurred; to explain, in some cases, that the standard of care was met; to provide reasonable compensation rapidly when it was not met; and to keep disputes from escalating into full-blown litigation. Several approaches are promising.23 Communication-and-resolution programs, in which health care institutions proactively disclose errors, apologize, and offer compensation before the patient files a claim, could reduce the numbers and costs of lawsuits and speed the process of resolution. So could expanded use of mediation.
Another, farther-reaching reform would be to replace litigation in the courts with an administrative compensation system, akin to workers compensation. Administrative compensation proposals typically suggest that patients should not have to prove that their providers were negligent, only that their injury could have been avoided in an optimal system of care. Although politically challenging, such proposals are appealing because trying to avoid the emotional distress of being labeled as negligent may drive a great deal of defensive behavior among physicians.
A final approach is to give care providers a strong defense to allegations of malpractice, known as a “safe harbor,” if they can show that they followed an applicable, well-accepted, evidence-based practice guideline. By promoting national, evidence-based standards of care, instead of holding physicians to the prevailing standard in their state or community, safe harbors also have the potential to reduce geographic variation in the provision of some services.33
Malpractice reform alone is unlikely to solve the problem of overuse of health care services, which has multiple drivers. Fee-for-service reimbursement provides a financial incentive to order more services. Patients’ demands for services may be hard to resist. Finally, physicians may be averse to the risk of missing something and harming the patient—not just the risk of being sued when such harm occurs. Human beings’ known tendency to overestimate the probability of rare but serious risks reinforces physicians’ tendency to take excessive precautions.
Although malpractice reform can’t root out all of these problems, it needs to be part of their solutions. Our study suggests that innovative reforms that address the underlying causes of defensive medicine have potential rewards far beyond their advantages for individual patients and clinicians—rewards that make them worth pursuing.
This study was funded by the National Institute for Health Care Reform. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Veterans Affairs.
- ↵Without those weights, the sample would be biased toward containing sicker patients.
- ↵To access the Appendix, click on the Appendix link in the box to the right of the article online.
- ↵The risk index used secondary data from the National Practitioner Data Bank, the Medical Liability Monitor, market share reports published by the National Association of Insurance Commissioners, and the American Medical Association’s Physician Masterfile. For details, see Note 6.
- ↵Specifically, age, sex, race, and “snowbird” status (defined as patients with at least 15 percent of their visits in a different census division than that of their main residential location).